Vers la compr\'ehension automatique de la parole bout-en-bout \`a moindre effort

TL;DR

This paper explores methods to reduce computational costs in end-to-end automatic speech understanding for French, demonstrating that cost-effective strategies can maintain high performance levels.

Contribution

It introduces and evaluates learning strategies that lower training costs without sacrificing state-of-the-art results on the MEDIA corpus.

Findings

Cost reduction is achievable with minimal performance loss.

Certain learning strategies outperform others in efficiency.

State-of-the-art performance maintained with reduced training effort.

Abstract

Recent advances in spoken language understanding benefited from Self-Supervised models trained on large speech corpora. For French, the LeBenchmark project has made such models available and has led to impressive progress on several tasks including spoken language understanding. These advances have a non-negligible cost in terms of computation time and energy consumption. In this paper, we compare several learning strategies aiming at reducing such cost while keeping competitive performances. The experiments are performed on the MEDIA corpus, and show that it is possible to reduce the learning cost while maintaining state-of-the-art performances.